About
Nearly three decades shipping production software have taught me that the interesting problems are rarely the ones written down. Most of my recent years have been spent embedded inside Tier‑1 banks and payments suppliers, where the work that matters lives in the quiet gap between a regulator's rule set and a system's actual behaviour. Lately that gap is where I have been putting LLM tooling, RAG pipelines and MCP servers to use, opening up domain problems that have stayed stubbornly manual for decades.
I arrive as a consulting engineer: I read the rulebooks, listen to the teams, and try to leave behind something the client can carry forward with or without me.
My day is split fairly evenly between the engineering floor and the management table — briefing senior leaders, working alongside mid‑level programme and delivery managers, and pairing with the engineers actually building the thing. Clear writing and the right questions tend to do as much of the work as the code does. The colleagues I enjoy most, at every level, are the ones who want a straight answer and are willing to learn something along the way.
The best parts of this craft are the hidden rules — the implicit constraints that never make it into a specification but that every seasoned engineer on a payments floor carries in their head. I take quiet pleasure in drawing those out, turning them into tests, tooling and documentation, and handing them on. Mentoring is not a thing I do alongside the delivery; it is what makes the delivery durable.
I'm a regular fixture at AI and tech‑startup gatherings around London — sometimes on the stage, more often in the room — sharing what I've learned about payments, testing, and the practical art of putting LLM tooling to work on real domain problems, and learning a great deal from the people I meet there.
Selected work
-
cbprstar.com
Realistic ISO 20022 test‑message generation, built from SWIFT scheme rule sets — available as both a web UI and an MCP server.
Payments · Live site
-
unicode_babel
A small PyPI package for generating random unicode characters — handy for fuzzing and test data.
Python · PyPI
-
grim_test
A GRIM‑test implementation in Python: spotting impossible statistical means in published research.
Python · Statistics
-
cribbage_scorer
A clean scoring engine for the classic card game — a hobby project, properly tested.
Python · Gaming
-
dwm
Experiments with a tiling window manager — small, sharp tools for a focused desktop.
TypeScript
-
pyiso20022
Python tooling for working with ISO 20022 payment messages.
Python · Payments